alt Hacker NewsIt is incredibly easy now to get an idea to the prototype stage, but making it production-ready still needs boring old software engineering skills. I know countless people who followed the "I'll vibe code my own business" trend, and a few of them did get pretty far, but ultimately not a single one actually launched. Anyone who has been doing this professionally will tell you that the "last step" is what takes the majority of time and effort.
Replies
> the "last step" is what takes the majority of time and effort
Having worked extensively with vibe-coded software, the main problem for me is that I have tuned-off from the ai-code, and I dont see any skin-in-the-game for me. This is dangerous because it becomes increasingly harder to root-cause and debug problems because that muscle is atrophying. use-it or lose-it applies to cognitive skills (coding/debugging). Now, I lean negatively to ai-code because, while it seduces us with fast progress in the first 80%, the end outcome is questionable in terms of quality. Finally, ai-coding encourages a prompt-and-test or trial-and-error approach to software engineering which is frustrating and those with experience would prefer to get it right by design.
Software engineering is not “coding” though.
Before AI for the last 8 or so years now first at a startup then working in consulting mostly with companies new to AWS or they wanted a new implementation, it’s been:
1. Gather requirements
2. Do the design
3. Present the design and get approval and make sure I didn’t miss anything
4. Do the infrastructure as code to create the architecture and the deployment pipeline
5. Design the schema and write the code
6. Take it through UAT and often go back to #4 or #5
7. Move it into production
8. Monitoring and maintenance.
#4 and #5 can be done easily with AI for most run of the mill enterprise SaaS implementations especially if you have the luxury of starting from the ground up “post AI”. This is something you could farm off to mid level ticket takers before AI.
I also experienced this with my personal projects. It was really easy to just workshop a new feature. I'd talk to claude and get a nice looking implementation spec. Then I'd pass it on to a coding agent which would get 80% there but the last 20% would actually take lot more time. In the meantime I'd workshop more and more features leading to an evergrowing backlog and an anxiety that an agent should be doing something otherwise I'm wasting time. I brought this completely on myself. I'm not building a business, nothing would happen if I just didn't implement another feature.
It's really thrown off some old adages. It's now "the first 90% takes 90% of the time, the last 10% takes the other 90,000,000% of the time."
Just doesn't have the same ring to it.
Exactly, there have been loads of tools over time to make software development easier - like Dreamweaver and Frontpage to build websites without coding, or low/no-code platforms to click and drag software together, or all frameworks ever, or libraries that solve issues that often take time - and I'm sure they've had a cumulative effect in developer productivity and / or software quality.
But there's not one tool there that triggered a major boost in output or number of apps / libraries / products created - unless I missed something.
Sure, total output has increased, especially since the early 2010's thanks to both Github becoming the social network of software development, and (arguably) Node / JS becoming one of the most popular languages/runtimes out there attracting a lot of developers to publish a lot of tools. But that's not down to productivity or output boosting developments.
> Anyone who has been doing this professionally will tell you that the "last step" is what takes the majority of time and effort.
That's true, but even the "last step" is being accelerated. The 10% that takes 90% of the time has itself been cut in half.
An example is turning debug logs and bug reports into bugfixes, and performance stats into infrastructure migrations.
The time required to analyze, implement, and deploy those has been reduced by a large amount.
It still needs to be coupled with software engineering skills - to decide between multiple solutions generated by an LLM, but the acceleration is significant.
Well… because it is not almost possible do it solo.
Code is just one part of puzzle. Add: Pricing, marketing and ads, invoicing, VAT, make really good onboarding, measure churn rate, do customer service…
A lot of vibe coders are solopreneurs. You have to be very consistent and disciplined to make final product that sells.
I agree 100%. Boring old software skills are part of what it took to "write" this DSL, complete with a fully featured LSP:
https://github.com/williamcotton/webpipe
https://github.com/williamcotton/webpipe-lsp
(lots of animated GIFs to show off the LSP and debugger!)
While I barely typed any of this myself I sure as heck read most of the generated code. But not all of it!
Of course you have to consider my blog to be "in production":
https://github.com/williamcotton/williamcotton.com/blob/main...
The reason I'm mentioning this project is because the article questions where all the AI apps are. Take a look at the git history of these projects and question if this would have been possible to accomplish in such a relatively short timeframe! Or maybe it's totally doable? I'm not sure. I knew nothing about quite a bit of the subsystems, eg, the Debug Adapter Protocol, before their implementation.
> Anyone who has been doing this professionally will tell you that the "last step" is what takes the majority of time and effort.
This is true, and I bet there are thousands of people who are in this stage right now - having gotten there far faster than they would have without Claude Code - which makes me predict that the point made in the article will not age well. I think it’s just a matter of a bit more time before the deluge starts, something on the order of six more months.
Even if you have the app, you get to start the fun adventure of marketing it and actually trying to grow the damn thing
Even beyond the engineering there are 100 other things to do.
I launched a vibe coded product a few months ago. I spent the majority of my time
- making sure the copy / presentation was effective on product website
- getting signing certificates (this part SUCKS and is expensive)
- managing release version binaries without a CDN (stupid)
- setting up LLC, website, domain, email, google search indexing, etc, etc
> not a single one actually launched.
I think this represents a fundamental misunderstanding of how these AI tools are used most effectively: not to write software but to directly solve the problem you were going to solve with software.
I used to not understand this and agreed with the "where is all the shovelware" comments, but now I've realized the real shift is not from automating software creation, but replacing the need for it in the first place.
It's clear that we're still awhile away from this being really understood and exploited. People are still confusingly building webapps that aren't necessary. Here's two, somewhat related, examples I've come across (I spend a lot of time on image/video generation in my free time): A web service that automatically creates "headshots" for you, and another that will automatically create TikTok videos for you.
I have bespoke AI versions of both of these I built myself in an afternoon, running locally, creating content for prices that simply can't be matched by anyone trying to build a SaaS company out of these ideas.
What people are thinking: "I know, I can use AI to build a SaaS startup the sells content!" But building a SaaS company still requires real software since it has to scale to multiple users and use cases. What more and more people are realizing is "I can created the content for basically free on my desktop, now I need to figure out how to leverage that content". I still haven't cracked the code for creating a rockstar TikTok channel, but it's not because I'm blocked on the content end.
Similarly I'm starting to see that we're still not thinking about how to reorganize software teams to maximally exploit AI. Right now I see lots of engineers doing software the old way with an AI powered exo-skeleton. We know what this results in: massive PRs that clog up the whole process, and small bugs that creep up later. So long as we divide labor into hyper focused roles, this will persist. What I'm increasingly seeing is that to leverage AI properly we need to re-think how these roles actually work, since now one person can be much responsible for a much larger surface area rather than just doing one thing (arguably) faster.
I launched a draw.io competitor to the point that it is in production, but there is little activity on the site as far as signups are concerned. Doesn't deliver enough business value.
> and a few of them did get pretty far, but ultimately not a single one actually launched.
Having done this professionally for a very, very long time, software engineers aren't particularly good at launching products.
Technology has drastically lowered the barriers to bring software products to customers, and AI is a continuation of that trend.
How much longer will this be true, though? With improving computer use, it may be possible in the next ~year or so that agents will be able to wire up infrastructure and launch to production.
all they did is annoy their friends and family by sharing their vibeslop app and asking for "feedback".
I really dont know how to respond to these requests. I am going to hide out and not talk to anyone till this fad passes.
Reminds of the trend where everyone was dj wanting you to listen their mixtrack they made on abbleton live
90% done, just the other 90% to do...
succinct and accurate.
[dead]
It's helping with that part too. I was able to configure a grafana stack with the help of claude for our ansible scripts.
> It is incredibly easy now to get an idea to the prototype stage
Yup. And for most purposes, that's enough. An app does not have to be productized and shipped to general audience to be useful. In fact, if your goal is to solve some specific problem for yourself, your friends/family, community or your team, then the "last step" you mention - the one that "takes majority of time and effort" - is entirely unnecessary, irrelevant, and a waste of time.
The productivity boost is there, but it's not measured because people are looking for the wrong thing. Products on the market are not solutions to problems, they're tools to make money. The two are correlated, because of bunch of obvious reasons (people need money, solving a problem costs money, people are happy to pay for solutions, etc.), but they're still distinct. AI is dropping the costs of "solving the problem" part, much more than that of "making a product", so it's not useful to use the lack of the latter as evidence of lack of the former.